Shorts on metal1 created when FILL_CELLS include decoupling cap cells.

[alexroach-rl]

Hi guys,

I am finding that OpenROAD can cause shorts on metal1 when the FILL_CELLS variable includes decoupling cap cells. It seems that it does not check for obstructions in the decap cell with respect to the existing routing before placing it in the layout.

I have noticed that the standard platforms that come with OpenROAD either set the min routing layer as metal2 or do not include decoupling caps in the FILL_CELLS variable. Is this a known issue?

[maliberty]

If we left out the fill cell you would have a well continuity error so I don't see how this is a valid use model. It seems you need to follow the approach of other platforms and use metal2 as the min layer with you cells.

[alexroach-rl]

Hi Matt,

Thanks for the reply. I understand your point, but it's still a bit surprising that it puts in the decap cells with metal1 obstructions over the metal1 routes instead of using standard filler cells. If it respected the metal1 routes you could still have metal1 as the min layer, but set the routing adjustment to limit the amount of metal1, so an adequate number of decap cells could still be used during filler insertion.

[rovinski]

@alexroach-rl I believe you are performing some action in the wrong order. You should always insert fill cells before performing routing. The filler cell insertion does not check obstructions, it only checks for continuity with the cell boundaries and that all holes are filled. The router is the tool that actually checks for obstructions and avoids them. You will find the same issue with commercial tools as well.

As an aside, using decap cells as primary filler cells is not a recommended methodology. This can lead to excessively high leakage power by overfilling with cells which have leaky capacitors. You should instead use insert_decap which will insert only enough decap to meet the target capacitance threshold.

[alexroach-rl]

@rovinski Thanks for the info. I am using the OpenROAD-flow-scripts, which inserts the filler cells after routing (Step 5_3). The insert_decap command does sound like a better approach, but it doesn't appear to be part of the flow-scripts yet. I may look into using it as a way to prevent overfilling with leaky capacitors as you said.

[maliberty]

@luis201420 does insert_decap check for m1 obs?

[luis201420]

@maliberty No, insert_decap doesn't check for m1 obs.
insert_decap works the same as fill cell, using the positions where there are no cells.